DESIGN PRINCIPLES

FOR
INTERCONNECTION OF NETWORKS
BY
A.Dhana Rekha P.Tripura Nagavalli
3rd CSE 3rd CSIT
GUDLAVALLERU ENGINEERING COLLEGE
GUDLAVALLERU

ABSTRACT
With the development of individual computer networks comes the need to
interconnect them. Network designers are faced with heterogeneity of networks just
they were previously faced with heterogeneity of computers within a single
network.This paper shows that similar structuring techniques,namely multiplexing,
switching, cascading, wrapping and layering, can be applied, and that a set of simple
principles can be derived which facilitate greatly the design of the interconnection of
computer networks.These simple principles are applied to the analysis of some typical
examples of network interconnection problems, in areas of addressing, routing, non-
equivalent communication services, error control,flow control and terminal
access.Similar principles could be applied to some unresolved issues in computer
network interconnection, such as congestion control or administrative functions. It is
finally claimed that final objective of network interconnection studies are to
determine the set of international standards which are required to make network
interconnection straightforward in the near future.
1. Introduction
Data processing is gradually evolving from its original model,to networking and
distributed processing.Computers have been linked into individual networks to satisfy
needs of individual organizations. Now, networks must be interconnected to cater to
inter-organizational relationships.Even though this requirement for interconnection of
computer networks was identified early, it is only recently that the problem has been
widely recognized..A set of simple rules can tremendously help to analyse specific
interconnection problems, as well as improve potential interconnectability of a
network through proper design choices.First question to be raised is "What is specific
network interconnection, as opposed to building single network ?".Basically, an
interconnected set of networks can be considered from an external (user's) point of
view and from an internal (designer's) point of view.From a user's viewpoint, an
interconnected set of networks is not different from a single network.In particular, two
identical networks can usually be integrated into a single bigger one. In addition, it is
essential to preserve freedom in the design of future computer networks,but still be
able to interconnect them with existing ones. In other words, the question is "how to
interconnect heterogeneous networks" rather than "how to build a world wide
homogeneous network".Before being faced with the constraint of interconnecting
heterogeneous networks, network designers have been faced with the problem of
interconnecting heterogeneous computers.The set of techniques developed for this
latter purpose can be used again, with slight adaptations, for interconnecting
heterogeneous networks.In this paper,we first briefly gave the set of techniques used
to design heterogeneous computer networks (section 2). In section (3),we mentioned
how these techniques can be applied to network interconnection and which principles
can be derived. This will then be illustrated with a number of concrete examples
(sections 4 to 9). We concluded by indicating (section 10) some features which should
be included in the design of any network to improve its interconnectability.
2. Structuring techniques
From experience , a few simple and powerful structuring techniques have merged
which now form the basis of any computer network architecture, namely :
multiplexing, switching, cascading, wrapping and layering. These basic techniques
are briefly reviewed in the following :
Multiplexing
In any network, many resources are concurrently shared among several users or more
generally among several activities. A multiplexing mechanism will take care of
distributing the resource to the various activities which need it (see figure 2-I). In
simple cases the multiplexing mechanism is a local (non-distributed) activity ; i.e., its
decision concern only local resources and are based on locally available information
(e.g., allocation of transmission lines to packets for transmission between adjacent
nodes).
 Figure 2-I : Multiplexing resources between activities
Switching
When one resource is shared among several activities,it must be able to identify which
activity is concerned with its successive actions,(see figure 2-2) and possibly deduce
where to forward a request for the following actions Switching implies interpretation
of addresses and routing of requests.

Figure 2-2 : Addressing actions

Cascading
Cascading consists in forming a linear string of entities (see figure 2-3) which forward
requests along the cascade (e.g., forwarding a packet along the path between source
and destination in a packet switching network). Cascading is the only way for
communication between entities which are not directly connected to each other.

Figure 2-3 : Cascading assembly

Wrapping
When functions performed by a set of cooperating entities are not exactly those
required by their users, a layer of entities can be added, wrapping the initial set (see
figure 2-4). Entities in this wrapping layer communicate through the initial set and
perform the additional or modified functions.
 Figure 2-4 : Wrapping
Layering
Layering is the technique by which most distributed systems are logically organized
(figure 2-5). Entities in the lowest layer are directly interconnected into a network and
provide communication services between entities in the next higher layer. These
entities,now provide services to entities in the layer above them. The layering
structure expands up to the highest layer.Layering can simply be viewed as the
recurrent application of wrapping.

a) A physical view of layering

b) A simplified logical view of layering

Figure 2-5:Layering assembly
Visibility
An essential concept underlying in structuring of computer networks is the concept of
visibility in any part of a network.The internal functioning of a part of the network is
not visible from outside of it.This concept of restricted visibility permits heterogeneity
to be reconciled with compatibility (i.e., common conventions between systems)
3. Architecture principles for network interconnection
3.1. Gateways
In general, networks are not compatible with each other and cannot be interconnected
directly. Intermediate gateways are required then (see figure 3-I).A gateway function
may be implemented in separate equipment connected to two or more networks, but it
may also be implemented in one or several pieces as additional modules in already
existing equipment.Now assume that the gateway functions are physically isolated.
The resulting structure is a network of networks (global network)interconnected by
gateways, as illustrated in figure 3-1.

.
Fig.3-1 : Networks interconnected by gateways
3.2. Cascading of services by gateways
The minimum function of a gateway is to forward information from one network to
another.A gateway is thus a place where individual network functions are cascaded.In
this subsection,we examine some of the implications of this cascading role of a
gateway on the general structure of the global network.Consider simple example of
the gateway between two networks A and B(in fig 3-I).We assume that each network
has a layered structure but that both structures are not identical (see figure 3-2).
 Figure 3-2 :Cascading by Gateway
The purpose of the gateway in figure 3-2 is to present to usersXandY the
concatenation of networksAandB as a single global networkA B, without intervening
in the cooperation between X and Y.In order for the gateway to be able to cascade the
two networks,services and cascaded by a gateway must be equivalent.Another
implication is that services and must be "cascadable".In order to match these
constraints,it is necessary to add a new layer to reach equivalent cascadable services
(figure 3-3). This new layer appears only in the user equipment and in the gateway,
thus wrapping (end-to-end) the original network without requiring any modification
to intermediate nodes in the network.

Figure 3.3: Adapting services to be equivalent

3.3. Multiplexing of gateways
A gateway must be shared among a number of internetwork activities .This should be
achieved by multiplexing gateway between the activities. Similarly interface between
gateway and each network must also be shared.There are cases where network
interface cannot be shared.Then original network will have to be wrapped (end-to-
end) in an additional layer which will provide interface multiplexing capability.
3.4. Switching by gateways
In the global network, gateways have to perform switching.If each individual network
is able to interpret addresses of all users of global network, the gateway has only to
decide to which network it should forward request. But, if an individual network is
able to interpret only its own address space, the gateway must be prepared to decide
for local address of final user.In this latter case, it is common practice to complement
individual network by wrapping it (end-to-end) into an additional layer which takes
care of handling the global address space.
3.5. Level of interconnection
The assumption made here is that these users have common protocols which allow
them to cooperate without the help of a gateway. These users may form the higher
level part of the global network - a common higher-levelarchitecture(see figure 3-4).

Fig 3.4 : Level of interconnection

3.6. Summary of interconnection principles
As outlined above, networks to be interconnected must exhibit levels of equivalent
services, to be possibly merged into a global network.If not,network(s) must be
modified, usually by wrapping it (end-to-end) in an additional layer which externally
exhibits required interconnection capability.The principles above are illustrated by a
series of examples in the following sections.
4. Addressing. and Routing
Being able to address a large population is a key aspect of network
interconnection.This section is intended to illustrate how switching can be performed
in such a global network.Packet network interconnection following the CATENET
approach, Public Data Network interconnectlon, as well as interconnection of private
networks with public networks, have been chosen as examples for the purpose of this
illustration.

The global network needs to provide a global address space(figure 4-3).This global
address space have the first part of the global address as an individual network while
the second part of the address identifies the user within that network(figure 4-2).

In the case where public networks offering virtual circuit services are interconnected
by means of the X.75 interface, addresses cover any user connected to any public
network in the world.In the case where a private network is interconnected with a
public network, by means of the X.25 interface, the private network is provided with a
single DTE address.Address extensions are required which can be performed by
wrapping public networks in an additional layer of addressing covering the
interconnection of public as well as private networks.
4. 2. Routing functions
The global network in figure 4-3 needs to perform routing functions.In the CATANET
approach,each node within each individual network is able to process both the first
part of all global addresses and the second part of global addresses of its own
users.Routing functions within each individual network forwards information closer
to its final destination.The routing function can easily be cascaded. (figure 4-4).

In the interconnection of public network,once an internetwork path has been

established,it cannot be changed.In the example of figure 4-5 below, the alternate path
through G2 and G3 cannot used once G1 and G4 have been chosen.
5. Equivalence of services
Interconnection of networks can only be performed at a level providing equivalent
services on either side. In addition these services must be cascadable.For example,
when a private network offering a datagram service is to be interconnected with a
public network providing virtual circuit services,the question of choosing appropriate
level of service where the networks can be interconnected needs to be studied.
(1) One can choose to interconnect them at the level of virtual circuit services (see
figure 5-I). In this case, a layer needs to be added wrapping the datagram network in
order to perform virtual circuit functions on top of the datagram service.In
practice,one access line to a public network cannot be shared by several addresses, an
additional layer needs also to be provided wrapping virtual circuitfunctions for
internetwork addressing and routing.
(2) The other alternative is to interconnect the two networks at the level of datagram
services (see figure 5-2). In this case, a layer needs to be added, wrapping the virtual
circuit network in order to provide datagram services on top of virtual circuits

6. Error control
Error control functions are generally based on transmitting, back to the source, control
messages,referring to blocks of information transmitted from a source to a destination.
Blocks are generally transmitted before previous blocks are acknowledged, in order to
optimize available bandwidth utilization despite a possibly long round trip delay.Error
control services provides two kinds of error protection.When interconnecting
datagram networks, error control services can be kept unchanged for the global
network (see figure 6-I).

When error-control is performed end-to-end on both networks, the level of service

may be preserved if the end-to-end significance of acknowledgements is also
preserved. In other words, gateways must forward acknowledgements transparently
(see figure 6-1).
 Figure 6-2 Passive gateway preserving end-to-end error control
. The choice made in X.75 is to choose as a common service not to consider the end-
to-end significance of acknowledgements on the network providing it. The gateways
between such networks will actually be considered as the end user (see figure 6-3).

7. Flow control
As for error control functions, networks performing flow control in a cascade of flow
control functions between adjacent nodes can easily be interconnected without
modifying their properties. This is the case when interconnecting virtual circuit
networks where flow control does not have end-to-end significance. (see figure 7-I).

End-to-end flow control services can be cascaded if the gateway simply translates
flow control information.
9. Other exaamples and issues
The principles developed in section (3) can also be applied to other problems related
to network interconnection, such as compatibilities of user interfaces and control
languages.They have also been applied to adapt closed systems to an open systems
environmentThis is in particular the case for congestion contro1 and administrative
functions in general, such as network supervision and maintenance, diagnostics,
accounting and billing. Nowadays, the most usual way to solve these problems is to
consider that the global network as such does not benefit from those functions which
are handled independently within each network.This is acceptable for initial usage of
a small number of networks, but would be intolerable in a few years with the expected
extensive usage of myriads of networks.
10. Conclusion
Interconnection of computer networks is a complex problem and largely still an open
question. However, it has been solved satisfactorily in a number of cases, permitting
partial interconnection .Experience shows that a set of simple rules can be applied to
analyze network interconnection problems. Of course, these simple principles are not
sufficient, and practical experience is still essential. It could reasonably be expected
that the same type of techniques could be applied to the remaining network
interconnection issues, but this still to be tried. The final objective of all present
studies and experiments in network interconnection should be to determine which
common properties networks must exhibit to make them readily interconnectable, and
to establish them as international standards.Common levels of services, expandability
of network addresses to global addresses, common layering structure, common
protocols on top of common services are such candidates for standardization.
11. References
(I) L. POUZIN, "Interconnection of packet switchirg network", IFIP WG 6.1 INWG
Note 42, 19 p., Oct. 1973.
(2) V. CERF, A. MC KENZIE, R. SCANTLEBURY, H. ZIMMERMANN, "Proposal
for an international end-to-end protocol"~ ACM Sigcom Comp. Com. Rev., vol. 6
(3) ISO, "Reference model of Open Systems Inter- connection",
ISO/TC97/SCI6/N227, 181 p. June 79
(4) L. POUZIN, H. ZIMMERMANN, "A tutorial on Protocols, Proc. IEEE, vol. 66,
(5) CCITT, "Recommendation X.25 : Interface between
DTE and DCE for terminals operating in the packet mode on public data networks",
Public Data Networks, Orange Book, vol. VIII.2, I.T.U. Geneva, Switzerland, 1977.
(6) A. MC KENZIE, "Some computer network interconnection issues", Proc. NCC'74,
AFIP Press, pp. 857-859, 1974.
(7) M. GLEN, J. LAWS, R. SCANTLEBURY, "Interconnection of packet switched
networks : theory and practice", Proc. EUROCOMP Communications Networks.
(8) P. HIGGINSON, A. HINCHLEY, "The problems of linking several networks with
a gateway computer", Proc. EUROCOMP Communications Networks, ONLINE, pp.
453-465, Sept. 1975.
(9) C.SUNSHINE,"Interconnection of computer networks",Computer Networks,vol1
(10) M. GLEN, "Network interconnection and protocol conversion", Proc.
COMNET'77, Budapest, Hungary, pp. 39-57, Oct.1977.
(11) CCITT, "Recommendations X.3, X.25, X.28, X.29 on packet switched data
transmission services", I.T.U., Geneva, Switzerland, 1977.
(12) V. CERF, P. KIRSTEIN, "Issues in Packet-Network Interconnection". Proc. IEEE
vol 66